Comment by KeplerBoy

Inference is already cheap enough. It's basically free for the quantities people actually want to read.

LLMs only become expensive if you let them generate 1000 lines of code every 2 minutes.